Ex-ante Impacts of Agricultural Insurance: Evidence from a Field Experiment in Mali Ghada Elabed* & Michael R Carter** *Mathematica Policy Research **University of California, Davis & NBER BASIS Assets & Market Access Research Program & I 4 Index Insurance Innovation Initiative http://basis.ucdavis.edu. Annual Global Development Conference, Casablanca 13 June 2015
Uninsured Risk Is Costly Risk is costly: Makes Households Poor when it leads them to adopt less risky, but lower returning activities Keeps Households Poor not only when it de-capitalizes them in the wake of a shock, but also when it leads them to accumulate unproductive 'buer' assets in anticipation of shocks Can insurance have real development impacts? Not just ex post smoothing eects (see previous work) But allowing farmers to ex ante prudentially invest more and increase their average incomes This presentation looks at the investment and income impacts of index insurance for cotton farmers in Mali
Logic of Index Insurance Conventional insurance (based on individual loss adjustment) has a dismal record Costly to verify losses for smallholders Moral hazard if do not/cannot reliably verify losses Adverse selection Index insurance does not pay based on (veried) individual losses, but instead based on a cheap to measure 'index' that is correlated with individual losses (e.g., average yields in a zone, or rainfall) Cuts costs Eliminates moral hazard & adverse selection
Cotton Production in Mali Most farmers are smallholders and grow a mix of subsistence crops and cotton Cotton is their main (and often only) source of cash Cotton is a protable, but risky crop Production organized in cooperatives Cotton is controlled by the Compagnie malienne pour le développement du textile (CMDT), a parastatal CMDT provides input loans and buys the harvest at a price announced before planting
Risk and Capital Constraints in Mali Farmers access credit via group loans: Amount of loan is on average 95, 000CFA/ha, and the net revenue from cotton is 105, 000CFA/ha If the cooperative yield falls below 750 kg/ha, loan repayment is tenuous P(yield < 750) = 10% Consequences of default are substantial (informal collateral) The collateral risk of default appears to discourage farmers from growing as much cotton as they otherwise mightclassic example of risk rationing Output can be taxed away to pay for others in the group
Dual Scale Area-yield Index Contract To address these issues, developed an area-yield insurance contract Can do this easily because monopsony buyer (CMDT) already measures area and output Insured unit is the cooperative (as it holds the joint liability debt for all farmers in the village) Payouts based on the average yield of the cooperative and of the ZPA grouping (an agglomeration of 10-15 village cooperatives) This dual scale area-yield contract has a low level of basis risk: Conditional on a loss, the probability of getting a payment is 80%, and the probability that net proceeds are less than the value of 750 kg of cotton per-hectare drops to 2%
The Mali Pilot Project In cooperation with PlaNet Guarantee implemented a randomized control trial for the 2011/12 year 87 cooperatives: 59 were randomly selected for treatment (oered insurance), 28 served as control An encouragement design: reduced the price of contract to 50%, 75%, or 100% of the actuarially fair premium Decisions to buy the insurance made in May 2011 (planting season) 30% of the treated cooperatives purchased the insurance contract
The Mali Pilot Project Note that the insurance purchase decision was a joint decision by co-op Creates the possibility that an individual farmer may not know s/he is insured (e.g., if missed meeting) Ex ante eects can of course only occur if farmers know they are insured In the analysis to follow, we will both look at the impacts of being insured (co-op purchased insurance) and also impact if farmer reports that s/he is insured.
Research design Unfortunately, we discovered aws in roll-out (not all treated co-ops were actually oered insurance) Fortunately, we had audit questions that allowed us to gure out what really happened But only 22.5% of the treated households believed they were insured (and 10% of non-treated households!) Results shown here use the audit-based reclassication of treatment and control areas Paper includes all resultssimilar estimated impacts but less precise Audit-adjusted treatment & control groups are balanced in terms of observable covariates Let's focus now on key outcome variables
Econometric Method The decision to buy insurance (and to know insured) are of course economically & econometrically endogenous To obtain unbiased estimates of the impacts of insurance purchase and farming knowing insured, we exploit our randomized controlled trial and use instrumental variable methods to recover standard local average treatment eects.
Impacts if Co-op Purchased Insurance
Impacts if Farmer Knows Insured
Conclusion Impact of insurance are substantial at the extensive margin: Area in cotton rose by 1.3 to 1.4 hectares (a 60% increase) Matching increases in loans and inputs No impacts on input intensity, nor any impact on reduction in other ag activity Output (and income) increases are estimated to be about 40%, but this gure is not signicant (noisy outcome measure) We thus see that insurance can have substantial development impact New pilot in Burkina Fasostay tuned!
Extra Material
Cumulative cotton area in 2011 Cumulatives: Coton area in 2011, below 3 ha Cumulatives: Coton area in 2011, above 3 ha 0.2.4.6.6.7.8.9 1 0 1 2 3 Area (ha) 0 5 10 15 20 Area (ha) cdf_insurance cdf_control cdf_insurance cdf_control
Prior Evidence on Risk TransferImpact of Index Insurance Maize producers in Ghana invest more when insured Randomly oered some farmers insurance at variable prices Other farmers oered a capital grant for purchasing inputs Found that farmers oered insurance: Expand area cultivated by 15% Increase input use by 40% Capital grants by themselves have little impact Source: Karlan et al. (2014). Agricultural Decisions after Relaxing Risk & Credit Constraints, Quarterly J of Econ.
Logic of Index Insurance Comparison of Index versus Conventional Insurance in Ecuador
Dual Scale Area-yield Index Contract Equivalently priced single & dual-scale contracts (Mali) Source: Elabed, Carter, Guirking & Bellemare (2014). Managing Basis Risk with Multi-scale Index Insurance Contracts, Agricultural Economics
Risk Rationing and Cotton Production Loans come with a binding joint liability clause Consequences of default are substantial (informal collateral) Survey in 2006: 32% of the farmers growing cotton had diculty with their loan repayment 38% had to sell their assets 4% sent one of their children to work for another farmer Some saw their credit line reduced and faced exclusion from the credit group The collateral risk of default appears to discourage farmers from growing as much cotton as they otherwise mightclassic example of risk rationing
Research question and Research Strategy There is modest but growing evidence that risk transfer (via insurance) & risk reduction (via stress tolerant varieties) has economically notable impacts Protecting current and future assets: Janzen and Carter (2014) Relaxing risk and capital constraints: Karlan et al (2014) Incentivizing technology adoption: Mobarak and Rosenzweig (2012) Use the remainder of our time to look at the investment and income impacts of insurance for cotton farmers in Mali We designed an insurance contract for cotton cooperatives in Mali We randomly oered insurance to 87 cotton cooperatives
Behavior of the Insured versus Uninsured These are NOT impact results
Risk Rationing and Cotton Production Loans come with a binding joint liability clause Consequences of default are substantial (informal collateral) Survey in 2006: 32% of the farmers growing cotton had diculty with their loan repayment 38% had to sell their assets 4% sent one of their children to work for another farmer Some saw their credit line reduced and faced exclusion from the credit group The collateral risk of default appears to discourage farmers from growing as much cotton as they otherwise mightclassic example of risk rationing
Dual Scale Area-yield Index Contract To address these issues, developed an area-yield insurance contract Can do this easily because monopsony buyer (CMDT) already measures area and output Insured unit is the cooperative (as it holds the joint liability debt for all farmers in the village) But at what scale set the trigger or strikepoint that determines payment? If take average across a larger area, basis risk increases If take average across too small an area, collusion and moral hazard possible
Dual Scale Area-yield Index Contract Rather than face the tradeo between a village level yield trigger (moral hazard) versus a yield trigger based on a larger ZPA grouping (an agglomeration of 10-15 village cooperatives), designed a dual scale contract Payouts based on the average yield of the cooperative AND of the ZPA Can think of the ZPA trigger as an audit ruleonly believe village yields are low due to nature if neighboring villages are also showing some signs of stress Conditional on a loss, the probability of getting a payment is 80%, and the probability that net proceeds are less than the value of 750 kg of cotton per-hectare drops to 2%
Logic of Index Insurance Conventional insurance (based on individual loss adjustment) has a dismal record Costly to verify losses for smallholders Moral hazard if do not/cannot reliably verify losses Adverse selection Index insurance does not pay based on (veried) individual losses, but instead based on a cheap to measure 'index' that is correlated with individual losses (e.g., average yields in a zone, or rainfall) Cuts costs Eliminates moral hazard & adverse selection
Baseline Balance on Key Variables
Econometric Method The decision to buy insurance (and to know insured) are of course economically & econometrically endogenous To obtain unbiased estimates of the impacts of insurance purchase and farming knowing insured, we exploit our randomized controlled trial and use instrumental variable methods to recover standard local average treatment eects: Y jc = α + βîjc + γx jc + ε jc where Y jc is the outcome variable of interest (e.g., yields) for farmer j in co-op c, X jc are a set of control variables (baseline wealth, experience, etc.), and Î jc is the rst stage estimate of whether or not individual jc is, or feels, insured.
Econometric Method We derive these rst stage estimates of insurance status by estimating two equations of this form: I jc = α + δ 1 T jc + δ 2 P jc + ε jc where I jc = 1 if individual jc is (or believes) that is insured, T jc is the randomly determined treatment variable (equals 1 if co-op was oered insurance) and P ic is a measure of the strike point penalty randomly imposed by the reinsurance company
Strikepoint Penalty Research team designed & then priced the contract under the assumption that yields for dierent co-ops in the same area are driven by a common parametric probability structure Structure was allowed to parametrically shift with each co-op's average long-term yields Resulted in a set of spatially stable prices
Strikepoint Penalty Reinsurance partner rejected this approach and priced each co-op separately using burn rates based on the short time series available on each co-op Resulted in often radical downward shift in strike points, with neighboring co-ops sometimes oered radically dierent contracts Also resulted in no contracts being oered to more than half the co-ops in pilot area Because these strike point dierences were driven by randomness (did one co-op happen to have an especially bad year in its time series, whereas its neighbor did not), a measure of this strike point penalty should serve as a statistically valid and strong instrument to explain insurance purchase
Strikepoint Penalty